sicabe-core/deploy/docker-compose.yml

# =============================================================================
# eBdesk SiCABE — Docker Compose Deployment
# Arsitektur 5 Lapisan: Ingestion | Storage | Processing | Serving | Governance
# Urutan Deployment sesuai Dependency Graph (bab 3.1 & 12.2):
#   1. Governance (IAM): Keycloak + OpenLDAP
#   2. Storage: MinIO → Nessie
#   3. Processing: Trino
#   4. Serving: StarRocks
#   5. Ingestion: Airflow
#   6. Governance (Lanjutan): OpenMetadata → Grafana
# =============================================================================

name: sicabe

networks:
  sicabe-network:
    name: sicabe-network
    driver: bridge
    ipam:
      config:
        - subnet: 172.28.0.0/16

volumes:
  traefik-acme:
  # ─── Database (Shared PostgreSQL — semua service) ─────────────────────
  postgres-data:

  # ─── Governance Layer ──────────────────────────────────────────────────
  keycloak-data:

  # ─── Identity (LDAP group provider — sumber grup untuk Trino/StarRocks) ─
  openldap-data:
  openldap-config:

  # ─── Storage Layer ─────────────────────────────────────────────────────
  minio-data:

  # ─── Processing Layer (coordinator & worker dipisah) ───────────────────
  trino-coordinator-data:
  trino-worker-data:

  # ─── Serving Layer (shared-data: FE + CN) ──────────────────────────────
  starrocks-fe-data:
  cn-cache:
  # ─── Serving Layer (Paralel: Dremio) ───────────────────────────────────
  dremio-data:

  # ─── Ingestion Layer ───────────────────────────────────────────────────
  airflow-logs:
  airflow-dags:
  # Config pipeline yang DIBANGKITKAN plugin OpenMetadata (DAG TestSuite/DQ).
  # Tanpa volume ini, config hilang saat container Airflow di-recreate sehingga
  # DAG yang tertinggal di airflow-dags gagal diimpor:
  #   ConfigurationError: Cannot open config file .../dag_generated_configs/<name>.json
  airflow-dags-configs:

  # ─── Governance Layer (Lanjutan) ───────────────────────────────────────
  openmetadata-data:
  # Index OpenSearch 3.x. Bersifat DERIVED — dibangun ulang OpenMetadata saat
  # re-index; index 2.x tidak dimigrasikan karena formatnya berbeda antar
  # versi mayor (2.x -> 3.x tidak kompatibel).
  opensearch-data:
  grafana-data:
  prometheus-data:

# ============================================================================
# PHASE 1 — Governance Layer (IAM): Keycloak
# Dependency: tidak ada (deploy pertama)
# ============================================================================
services:

  postgres:
    image: postgres:15
    container_name: sicabe-postgres
    restart: unless-stopped
    networks:
      - sicabe-network
    volumes:
      - postgres-data:/var/lib/postgresql/data
      # Init script: create databases & users untuk Keycloak, Nessie, Airflow, OpenMetadata
      - ./config/postgres/init:/docker-entrypoint-initdb.d:ro
    environment:
      # Superuser (digunakan init script untuk membuat database lain)
      POSTGRES_DB: keycloak
      POSTGRES_USER: sicabe_admin
      POSTGRES_PASSWORD: ${POSTGRES_PASSWORD:?err}
      # Password untuk masing-masing service database (dibaca oleh init script)
      KEYCLOAK_DB_PASSWORD: ${KEYCLOAK_DB_PASSWORD:?err}
      NESSIE_DB_PASSWORD: ${NESSIE_DB_PASSWORD:?err}
      AIRFLOW_DB_PASSWORD: ${AIRFLOW_DB_PASSWORD:?err}
      OPENMETADATA_DB_PASSWORD: ${OPENMETADATA_DB_PASSWORD:?err}
    healthcheck:
      test: ["CMD-SHELL", "pg_isready -U sicabe_admin -d keycloak"]
      interval: 10s
      timeout: 5s
      retries: 5
      start_period: 30s
    # ⚠️ Port 5432 TIDAK di-expose ke host — hanya diakses internal (postgres:5432)
    # Untuk debugging dari host, gunakan: docker exec -it sicabe-postgres psql -U sicabe_admin -d keycloak
    profiles:
      - phase1
      - phase2
      - phase3
      - phase4
      - phase5
      - phase6
      - all

  keycloak:
    image: quay.io/keycloak/keycloak:${KEYCLOAK_VERSION:-26.7}
    container_name: sicabe-keycloak
    labels:
      - "traefik.enable=true"
      - "traefik.http.routers.kc.rule=Host(`kc.${CLUSTER_DOMAIN:-localhost}`)"
      - "traefik.http.routers.kc.entrypoints=websecure"
      - "traefik.http.routers.kc.tls=true"
      - "traefik.http.routers.kc.tls.certresolver=le"
      - "traefik.http.routers.kc.tls.domains[0].main=${CLUSTER_DOMAIN:-localhost}"
      - "traefik.http.routers.kc.tls.domains[0].sans=*.${CLUSTER_DOMAIN:-localhost}"
      - "traefik.http.services.kc.loadbalancer.server.port=8080"
    restart: unless-stopped
    networks:
      - sicabe-network
    depends_on:
      postgres:
        condition: service_healthy
    volumes:
      - keycloak-data:/opt/keycloak/data
      # Realm import akan dilakukan oleh bootstrap.sh setelah Keycloak siap
    environment:
      KC_DB: postgres
      KC_DB_URL: jdbc:postgresql://${POSTGRES_HOST:-postgres}:${POSTGRES_INTERNAL_PORT:-5432}/keycloak
      KC_DB_USERNAME: keycloak
      KC_DB_PASSWORD: ${KEYCLOAK_DB_PASSWORD:?err}
      # Keycloak 26.x (hostname v2): KC_HOSTNAME = URL PUBLIK LENGKAP (skema+host+port).
      # Inilah yang menentukan issuer OIDC, dokumen discovery, dan redirect login.
      #
      # JANGAN pakai KC_HOSTNAME_URL: opsi hostname v1 itu DIHAPUS di Keycloak 26.0.0,
      # jadi memasangnya tidak berefek apa pun (dulu sempat dipasang di sini sehingga
      # KEYCLOAK_PUBLIC_URL seolah berfungsi padahal tidak). KC_HOSTNAME_BACKEND_URL
      # juga bukan opsi Keycloak sama sekali.
      KC_HOSTNAME: ${KEYCLOAK_PUBLIC_URL:-http://localhost:${KEYCLOAK_PORT:-12080}}
      # Endpoint server-ke-server (token/jwks) mengikuti Host request, bukan URL publik.
      # Dengan ini container lain tetap bisa memakai nama internal keycloak:8080
      # sementara browser memakai alamat publik — tanpa perlu dua nilai issuer.
      KC_HOSTNAME_BACKCHANNEL_DYNAMIC: "true"
      KC_HTTP_ENABLED: "true"
      # Cache/klaster Infinispan. Default `ispn` memakai registry DI DATABASE, dan
      # pemeriksaan "Keycloak cluster health check" akan DOWN bila tabel `jgroups` masih
      # memuat nama cluster dari deployment lama di DB yang sama.
      # PENTING: container yang unhealthy DIBUANG dari routing oleh Traefik (v3:
      # "Filtering unhealthy or starting container"), sehingga Keycloak bisa hilang dari
      # ingress tanpa error apa pun. Untuk instalasi SATU Keycloak, `local` mematikan
      # klaster sekaligus menghapus pemeriksaan itu.
      KC_CACHE: ${KEYCLOAK_CACHE:-ispn}
      # KC_PROXY_HEADERS hanya bila ADA reverse proxy di depan (Traefik, tls.mode != none).
      # Tanpa proxy, memasangnya membuat Keycloak memercayai header X-Forwarded-* dari
      # siapa pun (host bisa dipalsukan klien). Generator mengisinya hanya saat ada proxy.
      KC_PROXY_HEADERS: ${KC_PROXY_HEADERS:-}
      # Keycloak 26.x: bootstrap admin. KEYCLOAK_ADMIN* deprecated (KC-SERVICES0110)
      # dan dihapus pada rilis berikutnya → pakai KC_BOOTSTRAP_ADMIN_*.
      # Nilai tetap dari env kita (KEYCLOAK_ADMIN / KEYCLOAK_ADMIN_PASSWORD di .env).
      KC_BOOTSTRAP_ADMIN_USERNAME: ${KEYCLOAK_ADMIN:-admin}
      KC_BOOTSTRAP_ADMIN_PASSWORD: ${KEYCLOAK_ADMIN_PASSWORD:?err}
      KC_METRICS_ENABLED: "true"
      KC_HEALTH_ENABLED: "true"
      KC_LOG_LEVEL: INFO
      # Mode dev: batasi heap JVM (tanpa ini Quarkus mengikuti RAM host → ±3.8 G)
      JAVA_OPTS_APPEND: "-Xmx768m"
      # Keycloak 27.x: health/metrics pindah ke management port 9000.
      # Disable management health agar tetap di port 8080 (backward compatible).
      KC_HTTP_MANAGEMENT_HEALTH_ENABLED: "false"
      # Database connection pool — pool kecil untuk menghindari timeout
      KC_DB_POOL_INITIAL_SIZE: 1
      KC_DB_POOL_MAX_SIZE: 5
      KC_DB_POOL_MIN_SIZE: 1
    # Mode dev: batasi memori container — Quarkus memakai MaxRAMPercentage=70
    # (bawaan) yang diukur dari RAM host bila tanpa limit. Heap efektif ≤700 MB.
    mem_limit: 1g
    # Build-time flags sebagai CLI arg
    command: ["start", "--db=postgres", "--health-enabled=true", "--metrics-enabled=true", "--db-pool-initial-size=1", "--db-pool-max-size=5", "--http-enabled=true", "--proxy-headers=xforwarded", "--http-management-health-enabled=false"]
    healthcheck:
      # Keycloak 24.x+ image tidak menyertakan curl/wget, gunakan bash /dev/tcp.
      # Cek endpoint /health/ready (bukan hanya port) agar sesuai Keycloak 27 behavior.
      test: ["CMD-SHELL", "{ printf 'GET /health/ready HTTP/1.0\\r\\n\\r\\n' >&0; grep -q 'HTTP/1.0 200'; } 0<>/dev/tcp/127.0.0.1/8080"]
      interval: 15s
      timeout: 10s
      retries: 10
      start_period: 60s
    ports:
      - "${KEYCLOAK_PORT:-12080}:8080"
    profiles:
      - phase1
      - phase2
      - phase3
      - phase4
      - phase5
      - phase6
      - all

# ============================================================================
# PHASE 1b — Identity: OpenLDAP (sumber grup untuk Trino & StarRocks)
# Dependency: tidak ada. LDAP = source of truth grup (user tetap di Keycloak).
# ============================================================================

  openldap:
    image: osixia/openldap:${OPENLDAP_VERSION:-1.5.0}
    container_name: sicabe-openldap
    restart: unless-stopped
    networks:
      - sicabe-network
    volumes:
      - openldap-data:/var/lib/ldap
      - openldap-config:/etc/ldap/slapd.d
    environment:
      LDAP_ORGANISATION: "SiCABE"
      LDAP_DOMAIN: "sicabe.local"
      LDAP_ADMIN_PASSWORD: ${LDAP_ADMIN_PASSWORD:?err}
      LDAP_CONFIG_PASSWORD: ${LDAP_CONFIG_PASSWORD:?err}
      LDAP_TLS: "false"
    healthcheck:
      test: ["CMD-SHELL", "ldapsearch -x -H ldap://localhost -b dc=sicabe,dc=local -D cn=admin,dc=sicabe,dc=local -w $${LDAP_ADMIN_PASSWORD} '(objectClass=*)' | grep -q 'dc=sicabe'"]
      interval: 20s
      timeout: 10s
      retries: 10
      start_period: 30s
    ports:
      - "${OPENLDAP_PORT:-12389}:389"
    profiles:
      - phase1
      - phase2
      - phase3
      - phase4
      - all

# ============================================================================
# PHASE 2 — Storage Layer: MinIO → Nessie
# Dependency: Keycloak (Phase 1) sudah running
# ============================================================================

  minio:
    image: ${MINIO_IMAGE:-quay.io/minio/minio:RELEASE.2025-09-07T16-13-09Z}
    container_name: sicabe-minio
    labels:
      - "traefik.enable=true"
      - "traefik.http.routers.minio.rule=Host(`minio.${CLUSTER_DOMAIN:-localhost}`)"
      - "traefik.http.routers.minio.entrypoints=websecure"
      - "traefik.http.routers.minio.tls=true"
      - "traefik.http.routers.minio.tls.certresolver=le"
      - "traefik.http.routers.minio.tls.domains[0].main=${CLUSTER_DOMAIN:-localhost}"
      - "traefik.http.routers.minio.tls.domains[0].sans=*.${CLUSTER_DOMAIN:-localhost}"
      - "traefik.http.services.minio.loadbalancer.server.port=9001"
    restart: unless-stopped
    networks:
      - sicabe-network
    volumes:
      - minio-data:/data
    environment:
      MINIO_ROOT_USER: ${MINIO_ROOT_USER:-minioadmin}
      MINIO_ROOT_PASSWORD: ${MINIO_ROOT_PASSWORD:?err}
      MINIO_BROWSER: "on"
      # URL yang dilihat BROWSER untuk console (SSO/redirect lewat reverse proxy).
      # Ini "current" (bukan deprecated; jangan tertukar dengan MINIO_BROWSER).
      # Dibiarkan kosong dulu = console memakai host internal → redirect tak terjangkau
      # saat ada proxy. Tanpa proxy (tls.mode none) nilainya sama dengan akses langsung.
      MINIO_BROWSER_REDIRECT_URL: ${MINIO_CONSOLE_PUBLIC_URL:-http://localhost:${MINIO_CONSOLE_PORT:-12901}}
      MINIO_PROMETHEUS_AUTH_TYPE: "public"
    command: server /data --console-address ":9001"
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:9000/minio/health/live"]
      interval: 15s
      timeout: 10s
      retries: 5
      start_period: 30s
    ports:
      - "${MINIO_PORT:-12900}:9000"
      - "${MINIO_CONSOLE_PORT:-12901}:9001"
    profiles:
      - phase2
      - phase3
      - phase4
      - all

  # ─── One-shot: buat bucket MinIO sebelum Nessie start (idempotent) ────────
  minio-init:
    image: ${MINIO_MC_IMAGE:-quay.io/minio/mc:RELEASE.2025-08-13T08-35-41Z}
    container_name: sicabe-minio-init
    restart: "no"
    networks:
      - sicabe-network
    depends_on:
      minio:
        condition: service_healthy
    entrypoint: ["/bin/sh", "-c"]
    command:
      - |
        mc alias set local http://${MINIO_HOST:-minio}:${MINIO_INTERNAL_PORT:-9000} "$${MINIO_ROOT_USER}" "$${MINIO_ROOT_PASSWORD}"
        for b in "$${MINIO_BUCKET_ICEBERG_DATA}"; do
          mc mb --ignore-existing "local/$${b}"
        done
        echo "[SiCABE] Bucket MinIO siap."
    environment:
      MINIO_ROOT_USER: ${MINIO_ROOT_USER:-minioadmin}
      MINIO_ROOT_PASSWORD: ${MINIO_ROOT_PASSWORD:?err}
      MINIO_BUCKET_ICEBERG_DATA: ${MINIO_BUCKET_ICEBERG_DATA:-iceberg-data}
    profiles:
      - phase2
      - phase3
      - phase4
      - all

  nessie:
    image: ${NESSIE_IMAGE:-ghcr.io/projectnessie/nessie:0.108.4-java}
    container_name: sicabe-nessie
    restart: unless-stopped
    networks:
      - sicabe-network
    depends_on:
      postgres:
        condition: service_healthy
      minio:
        condition: service_healthy
      minio-init:
        condition: service_completed_successfully
    environment:
      # ── Version store: JDBC2 di PostgreSQL (ikut konvensi shared Postgres) ──
      NESSIE_VERSION_STORE_TYPE: "JDBC2"
      NESSIE_VERSION_STORE_PERSIST_JDBC_DATASOURCE: "postgresql"
      QUARKUS_DATASOURCE_POSTGRESQL_DB_KIND: "postgresql"
      QUARKUS_DATASOURCE_POSTGRESQL_ACTIVE: "true"
      QUARKUS_DATASOURCE_POSTGRESQL_DEVSERVICES_ENABLED: "false"
      QUARKUS_DATASOURCE_POSTGRESQL_JDBC_URL: "jdbc:postgresql://${POSTGRES_HOST:-postgres}:${POSTGRES_INTERNAL_PORT:-5432}/nessie?currentSchema=nessie"
      QUARKUS_DATASOURCE_POSTGRESQL_USERNAME: "nessie"
      QUARKUS_DATASOURCE_POSTGRESQL_PASSWORD: "${NESSIE_DB_PASSWORD:?err}"

      # ── Iceberg REST catalog + warehouse di MinIO (S3 server-side) ──
      # Warehouse = bucket iceberg-data (konvensi base) + prefix warehouse/.
      NESSIE_CATALOG_DEFAULT_WAREHOUSE: "warehouse"
      NESSIE_CATALOG_WAREHOUSES_WAREHOUSE_LOCATION: "s3://${MINIO_BUCKET_ICEBERG_DATA:-iceberg-data}/warehouse"
      NESSIE_CATALOG_SERVICE_S3_DEFAULT_OPTIONS_ENDPOINT: "${S3_ENDPOINT:-http://minio:9000}"
      NESSIE_CATALOG_SERVICE_S3_DEFAULT_OPTIONS_REGION: "us-east-1"
      NESSIE_CATALOG_SERVICE_S3_DEFAULT_OPTIONS_AUTH_TYPE: "STATIC"
      NESSIE_CATALOG_SERVICE_S3_DEFAULT_OPTIONS_PATH_STYLE_ACCESS: "true"
      # Kredensial S3 via Quarkus secret store. Nama secret memakai TITIK
      # (bukan hyphen) — hyphen tidak ter-decode benar oleh Quarkus env mapping
      # (temuan eksperimen nessie-dremio).
      NESSIE_CATALOG_SERVICE_S3_DEFAULT_OPTIONS_ACCESS_KEY: "urn:nessie-secret:quarkus:minio.creds"
      MINIO_CREDS_NAME: "${MINIO_ROOT_USER:-minioadmin}"
      MINIO_CREDS_SECRET: "${MINIO_ROOT_PASSWORD:?err}"
    # Mode dev: batasi memori container. Image Nessie memakai
    # -XX:MaxRAMPercentage=80 (bawaan) yang diukur dari RAM host → tanpa limit
    # heap-nya bisa ±12 GB. Dengan limit 1g, heap efektif ±800 MB.
    mem_limit: 1g
    healthcheck:
      # Image Nessie tanpa curl/wget → probe /dev/tcp ke management port 9000.
      test: ["CMD", "bash", "-c", "exec 3<>/dev/tcp/localhost/9000 && printf 'GET /q/health/ready HTTP/1.0\\r\\n\\r\\n' >&3 && cat <&3 | grep -q '200 OK'"]
      interval: 15s
      timeout: 10s
      retries: 15
      start_period: 60s
    ports:
      - "${NESSIE_PORT:-12120}:19120"
    profiles:
      - phase2
      - phase3
      - phase4
      - all


# ============================================================================
# PHASE 3 — Processing Layer: Trino + OPA (Authorization)
# Dependency: Storage (MinIO, Nessie) + Keycloak sudah running
# ============================================================================

  opa:
    image: openpolicyagent/opa:${OPA_VERSION:-1.20.2}
    container_name: sicabe-opa
    restart: unless-stopped
    networks:
      - sicabe-network
    # Decision point authz Trino: policy Rego + data ACL lokal (deny-by-default).
    # TODO Fase 7: pin versi OPA stabil.
    command: ["run", "--server", "--log-level=info", "--addr=:8181", "/policies", "/data/data.json"]
    volumes:
      - ./config/opa/policies:/policies:ro
      - ./config/opa/data.json:/data/data.json:ro
    # Tanpa healthcheck: image OPA tidak menyediakan shell/curl.
    # Kesiapan diverifikasi dari host (bootstrap/verify) via decision endpoint.
    ports:
      - "${OPA_PORT:-12181}:8181"
    profiles:
      - phase3
      - phase4
      - all

  # ── Trino — coordinator & worker DIPISAH sejak awal ───────────────────────
  # Agar tier R11 bersih: `trino-coordinator` = tier `coord` (koordinasi, auth,
  # client-facing), `trino-worker` = tier `engine` (eksekusi task, scale-out).
  # Single-host: keduanya di host yang sama. Multi-host: coordinator → host
  # `coord`, worker → host `engine`. Klien hanya melihat HTTPS coordinator.
  trino-coordinator:
    image: trinodb/trino:${TRINO_VERSION:-483}
    container_name: sicabe-trino-coordinator
    labels:
      - "traefik.enable=true"
      - "traefik.http.routers.trino.rule=Host(`trino.${CLUSTER_DOMAIN:-localhost}`)"
      - "traefik.http.routers.trino.entrypoints=websecure"
      - "traefik.http.routers.trino.tls=true"
      - "traefik.http.routers.trino.tls.certresolver=le"
      - "traefik.http.routers.trino.tls.domains[0].main=${CLUSTER_DOMAIN:-localhost}"
      - "traefik.http.routers.trino.tls.domains[0].sans=*.${CLUSTER_DOMAIN:-localhost}"
      - "traefik.http.services.trino.loadbalancer.server.port=8443"
      - "traefik.http.services.trino.loadbalancer.server.scheme=https"
      - "traefik.http.services.trino.loadbalancer.serverstransport=trino-insecure@file"
    hostname: trino-coordinator
    restart: unless-stopped
    networks:
      - sicabe-network
    depends_on:
      nessie:
        condition: service_healthy
      minio:
        condition: service_healthy
      opa:
        condition: service_started
    volumes:
      - trino-coordinator-data:/data/trino
      # Konfigurasi coordinator (HTTPS + OAuth2/Basic, discovery.uri)
      - ./config/trino/config-coordinator.properties:/etc/trino/config.properties:ro
      # Access Control via OPA (bab 4.3.1.6)
      - ./config/trino/access-control.properties:/etc/trino/access-control.properties:ro
      # Konfigurasi konektor Iceberg dengan Nessie (bab 4.3.1.3)
      - ./config/trino/catalog/iceberg.properties:/etc/trino/catalog/iceberg.properties:ro
      # Konfigurasi resource groups (bab 4.3.1.5)
      - ./config/trino/resource-groups.properties:/etc/trino/resource-groups.properties:ro
      - ./config/trino/resource-groups.json:/etc/trino/resource-groups.json:ro
      # Konfigurasi JVM tuning — coordinator (heap 512m, tidak mengeksekusi task)
      - ./config/trino/jvm-coordinator.config:/etc/trino/jvm.config:ro
      # Konfigurasi log
      - ./config/trino/log.properties:/etc/trino/log.properties:ro
      # LDAP GroupProvider config (grup dari OpenLDAP)
      - ./config/trino/group-provider.properties:/etc/trino/group-provider.properties:ro
      # Password authenticator (HTTP Basic) untuk service account
      - ./config/trino/password-authenticator.properties:/etc/trino/password-authenticator.properties:ro
      - ./config/trino/password.db:/etc/trino/password.db:ro
      # Keystore TLS untuk Web UI OAuth2 (OAuth2 Web UI wajib HTTPS)
      - ./config/trino/keystore.p12:/etc/trino/keystore.p12:ro
    environment:
      TRINO_HOME: /etc/trino
      # Static catalog management — load dari etc/catalog/*.properties
      CATALOG_MANAGEMENT: static

      # Native S3/MinIO credentials (native FS, Trino 481+)
      AWS_ACCESS_KEY_ID: ${MINIO_ROOT_USER:-minioadmin}
      AWS_SECRET_ACCESS_KEY: ${MINIO_ROOT_PASSWORD:?err}
      AWS_REGION: ${S3_REGION:-us-east-1}
      AWS_ENDPOINT_URL: ${S3_ENDPOINT:-http://minio:9000}
      # Dipakai oleh ${ENV:...} di catalog/iceberg.properties (bisa diarahkan ke S3 existing)
      MINIO_ROOT_USER: ${MINIO_ROOT_USER:-minioadmin}
      MINIO_ROOT_PASSWORD: ${MINIO_ROOT_PASSWORD:?err}
      MINIO_BUCKET_ICEBERG_DATA: ${MINIO_BUCKET_ICEBERG_DATA:-iceberg-data}
      S3_ENDPOINT: ${S3_ENDPOINT:-http://minio:9000}
      S3_REGION: ${S3_REGION:-us-east-1}
      S3_ACCESS_KEY: ${S3_ACCESS_KEY:-${MINIO_ROOT_USER:-minioadmin}}
      S3_SECRET_KEY: ${S3_SECRET_KEY:-${MINIO_ROOT_PASSWORD:?err}}
      NESSIE_URL: ${NESSIE_URL:-http://nessie:19120}
      # OAuth2 client credential untuk Keycloak
      TRINO_CLIENT_SECRET: ${TRINO_CLIENT_SECRET:?err}
      # URL publik Keycloak — dipakai config.properties lewat ${ENV:KEYCLOAK_PUBLIC_URL}
      # untuk issuer + auth-url. Trino GAGAL START bila variabel ini tidak ada (bukan
      # dianggap kosong), jadi sengaja diberi default eksplisit di sini.
      KEYCLOAK_PUBLIC_URL: ${KEYCLOAK_PUBLIC_URL:-http://localhost:${KEYCLOAK_PORT:-12080}}
      # Di belakang Traefik (tls.mode edge/full) nilai ini `true`. Wajib ada: Trino gagal
      # start bila variabel yang dirujuk config.properties tidak tersedia.
      TRINO_PROCESS_FORWARDED: ${TRINO_PROCESS_FORWARDED:-false}
      # LDAP GroupProvider: DN + password akun BIND (OpenLDAP / IAM existing).
      # LDAP_ADMIN_DN = DN akun bind — boleh akun READ-ONLY, bukan harus admin. Di-expand
      # oleh group-provider.properties lewat ${ENV:LDAP_ADMIN_DN}. Wajib ada: Trino gagal
      # start bila variabel yang dirujuk config tidak tersedia.
      LDAP_ADMIN_DN: ${LDAP_ADMIN_DN:-cn=admin,dc=sicabe,dc=local}
      LDAP_ADMIN_PASSWORD: ${LDAP_ADMIN_PASSWORD:?err}
      LDAP_URL: ${LDAP_URL:-ldap://openldap:389}
      # Port internal Keycloak untuk config Trino (config-coordinator.properties
      # memakai ${ENV:KEYCLOAK_INTERNAL_PORT}) — wajib ada di container.
      KEYCLOAK_INTERNAL_PORT: ${KEYCLOAK_INTERNAL_PORT:-8080}
      # Host Keycloak untuk token-url/jwks-url (server-ke-server). Multi-host dgn Keycloak
      # EKSTERNAL: ini = ALAMAT host pemilik (mis. 100.93.54.108) — bukan nama container
      # `keycloak` yang tak bisa di-resolve → penukaran token gagal ("could not be verified").
      KEYCLOAK_HOST: ${KEYCLOAK_HOST:-keycloak}
      # Kunci HMAC tetap untuk state OAuth2 Trino (default Trino acak tiap start).
      TRINO_OAUTH_STATE_KEY: ${TRINO_OAUTH_STATE_KEY:-trino-oauth-state-dev-only}
    healthcheck:
      # Trino internal port adalah 8080 (http-server.http.port)
      test: ["CMD", "curl", "-f", "http://localhost:8080/v1/info"]
      interval: 15s
      timeout: 10s
      retries: 10
      start_period: 60s
    ports:
      # Host 12443 → Container 8443 (HTTPS: OAuth2 Web UI + Basic service account)
      - "${TRINO_HTTPS_PORT:-12443}:8443"
      # HTTP polos TIDAK dilayani (http-server.authentication.allow-insecure-over-http=false)
      # dan tidak ada port HTTP yang di-expose — hanya HTTPS di atas.
    profiles:
      - phase3
      - all

  # ─── trino-jmx-exporter: saturasi Trino (JMX) → Prometheus (M3 observabilitas) ─
  # Sidecar yang membaca JMX coordinator via RMI (lihat jvm-coordinator.config) dan
  # menyajikannya sebagai metrik Prometheus di :9404. Satu tier dengan trino-coordinator
  # (`coord`) agar koneksi JMX tetap lokal; lihat lib/cluster/inventory.py.
  trino-jmx-exporter:
    image: bitnami/jmx-exporter:${JMX_EXPORTER_VERSION:-latest}
    container_name: sicabe-trino-jmx-exporter
    restart: unless-stopped
    networks:
      - sicabe-network
    depends_on:
      trino-coordinator:
        condition: service_started
    volumes:
      - ./config/trino/jmx-exporter.yml:/config/config.yml:ro
    # Entrypoint image: `java -jar jmx_prometheus_standalone.jar` → arg = <port> <config>.
    command: ["9404", "/config/config.yml"]
    profiles:
      - phase3
      - all

  # Worker: TIDAK ada `ports:` (sengaja) — klien tidak boleh menyentuhnya.
  # Ia mendaftar ke coordinator via discovery.uri (config-worker.properties).
  trino-worker:
    image: trinodb/trino:${TRINO_VERSION:-483}
    container_name: sicabe-trino-worker
    hostname: trino-worker
    restart: unless-stopped
    networks:
      - sicabe-network
    depends_on:
      trino-coordinator:
        condition: service_healthy
    volumes:
      - trino-worker-data:/data/trino
      # Konfigurasi worker (coordinator=false; tanpa HTTPS/auth klien)
      - ./config/trino/config-worker.properties:/etc/trino/config.properties:ro
      # Worker juga membaca katalog yang sama (data di Snowflake-like remote FS)
      - ./config/trino/catalog/iceberg.properties:/etc/trino/catalog/iceberg.properties:ro
      # Konfigurasi JVM tuning — worker (heap 1G, mengeksekusi task)
      - ./config/trino/jvm-worker.config:/etc/trino/jvm.config:ro
      # Konfigurasi log
      - ./config/trino/log.properties:/etc/trino/log.properties:ro
    environment:
      TRINO_HOME: /etc/trino
      # Static catalog management — load dari etc/catalog/*.properties
      CATALOG_MANAGEMENT: static

      # Env yang dipakai ${ENV:...} di catalog/iceberg.properties → wajib di worker juga
      AWS_ACCESS_KEY_ID: ${MINIO_ROOT_USER:-minioadmin}
      AWS_SECRET_ACCESS_KEY: ${MINIO_ROOT_PASSWORD:?err}
      AWS_REGION: ${S3_REGION:-us-east-1}
      AWS_ENDPOINT_URL: ${S3_ENDPOINT:-http://minio:9000}
      MINIO_ROOT_USER: ${MINIO_ROOT_USER:-minioadmin}
      MINIO_ROOT_PASSWORD: ${MINIO_ROOT_PASSWORD:?err}
      MINIO_BUCKET_ICEBERG_DATA: ${MINIO_BUCKET_ICEBERG_DATA:-iceberg-data}
      S3_ENDPOINT: ${S3_ENDPOINT:-http://minio:9000}
      S3_REGION: ${S3_REGION:-us-east-1}
      S3_ACCESS_KEY: ${S3_ACCESS_KEY:-${MINIO_ROOT_USER:-minioadmin}}
      S3_SECRET_KEY: ${S3_SECRET_KEY:-${MINIO_ROOT_PASSWORD:?err}}
      NESSIE_URL: ${NESSIE_URL:-http://nessie:19120}
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8080/v1/info"]
      interval: 15s
      timeout: 10s
      retries: 10
      start_period: 60s
    profiles:
      - phase3
      - all


# ============================================================================
# PHASE 4 — Serving Layer: StarRocks (shared-data: FE + Compute Node)
# Dependency: Storage (MinIO, Nessie) + Keycloak + OpenLDAP sudah running
# ============================================================================

  starrocks-fe:
    image: starrocks/fe-ubuntu:${STARROCKS_VERSION:-4.0.13}
    container_name: sicabe-starrocks-fe
    # Hostname tetap: identitas replika BDB FE memakai hostname; recreate
    # container TANPA hostname tetap → identitas stale → FE stuck INIT/UNKNOWN.
    hostname: sicabe-starrocks-fe
    restart: unless-stopped
    networks:
      - sicabe-network
    depends_on:
      nessie:
        condition: service_healthy
      minio:
        condition: service_healthy
      keycloak:
        condition: service_healthy
      openldap:
        condition: service_healthy
    volumes:
      - starrocks-fe-data:/opt/starrocks/fe
      # fe.conf: run_mode=shared_data + cloud_native_storage_type=S3 (WAJIB sebelum first start)
      - ./config/starrocks/fe.conf:/opt/starrocks/fe/conf/fe.conf:ro
    environment:
      STARROCKS_ROOT_PASSWORD: ${STARROCKS_ROOT_PASSWORD:?err}
      AWS_EC2_METADATA_DISABLED: "true"
      # Kredensial S3/MinIO untuk external catalog (Iceberg via Nessie).
      # Pakai S3_* dengan fallback saat-pakai ke MINIO_ROOT_* — konsisten dengan
      # Trino & Nessie, dan benar saat memakai S3 existing (§2.9/§2.13).
      AWS_ACCESS_KEY_ID: ${S3_ACCESS_KEY:-${MINIO_ROOT_USER:-minioadmin}}
      AWS_SECRET_ACCESS_KEY: ${S3_SECRET_KEY:-${MINIO_ROOT_PASSWORD:?err}}
      AWS_REGION: ${S3_REGION:-us-east-1}
      AWS_ENDPOINT_URL: ${S3_ENDPOINT:-http://minio:9000}
    command: ["bash", "/opt/starrocks/fe/bin/start_fe.sh", "--host_type", "FQDN"]
    healthcheck:
      test: ["CMD-SHELL", "mysql -u root -p$${STARROCKS_ROOT_PASSWORD} -h 127.0.0.1 -P 9030 -e \"SHOW FRONTENDS\" 2>/dev/null | grep -q \"Alive\" | head -1"]
      interval: 20s
      timeout: 15s
      retries: 15
      start_period: 120s
    ports:
      - "${STARROCKS_FE_PORT:-12930}:9030"
      - "${STARROCKS_HTTP_PORT:-12030}:8030"
    profiles:
      - phase4
      - all

  starrocks-cn:
    image: starrocks/cn-ubuntu:${STARROCKS_VERSION:-4.0.13}
    container_name: sicabe-starrocks-cn
    hostname: sicabe-starrocks-cn
    restart: unless-stopped
    networks:
      - sicabe-network
    depends_on:
      starrocks-fe:
        condition: service_started
    volumes:
      - cn-cache:/opt/starrocks/cn/cache
      - ./config/starrocks/cn.conf:/opt/starrocks/cn/conf/cn.conf:ro
    environment:
      HOST_TYPE: FQDN
      STARROCKS_ROOT_PASSWORD: ${STARROCKS_ROOT_PASSWORD:?err}
      AWS_EC2_METADATA_DISABLED: "true"
      # Sama dengan FE: pakai S3_* dengan fallback saat-pakai (benar untuk S3 existing)
      AWS_ACCESS_KEY_ID: ${S3_ACCESS_KEY:-${MINIO_ROOT_USER:-minioadmin}}
      AWS_SECRET_ACCESS_KEY: ${S3_SECRET_KEY:-${MINIO_ROOT_PASSWORD:?err}}
      AWS_REGION: ${S3_REGION:-us-east-1}
      AWS_ENDPOINT_URL: ${S3_ENDPOINT:-http://minio:9000}
      S3_ENDPOINT: ${S3_ENDPOINT:-http://minio:9000}
      S3_USE_PATH_STYLE_ENDPOINT: "true"
    command:
      - /bin/bash
      - -c
      - |
        ulimit -n 65535
        # 1) Tunggu FE siap, lalu pastikan root FE berpassword.
        for i in $$(seq 1 30); do
          if mysql --connect-timeout 3 -h ${STARROCKS_FE_HOST:-starrocks-fe} -P 9030 -u root -e "SELECT 1" >/dev/null 2>&1; then
            mysql --connect-timeout 3 -h ${STARROCKS_FE_HOST:-starrocks-fe} -P 9030 -u root \
              -e "SET PASSWORD FOR 'root' = PASSWORD('$${STARROCKS_ROOT_PASSWORD}')" >/dev/null 2>&1 \
              || mysql --connect-timeout 3 -h ${STARROCKS_FE_HOST:-starrocks-fe} -P 9030 -u root \
              -e "ALTER USER 'root'@'%' IDENTIFIED BY '$${STARROCKS_ROOT_PASSWORD}'" >/dev/null 2>&1 || true
            echo "[SiCABE] FE siap; root password diset (attempt $$i)."
            break
          fi
          echo "[SiCABE] Menunggu FE siap... (attempt $$i/30)"
          sleep 5
        done
        # 2) Daftarkan CN ini ke FE (shared-data → ADD COMPUTE NODE, bukan BACKEND).
        if mysql --connect-timeout 3 -h ${STARROCKS_FE_HOST:-starrocks-fe} -P 9030 -u root -p"$${STARROCKS_ROOT_PASSWORD}" \
             -e 'SHOW COMPUTE NODES;' 2>/dev/null | grep -q "${STARROCKS_CN_HOST:-starrocks-cn}"; then
          echo "[SiCABE] CN sudah terdaftar."
        else
          for i in $$(seq 1 20); do
            if mysql --connect-timeout 3 -h ${STARROCKS_FE_HOST:-starrocks-fe} -P 9030 -u root -p"$${STARROCKS_ROOT_PASSWORD}" \
              -e 'ALTER SYSTEM ADD COMPUTE NODE "${STARROCKS_CN_HOST:-starrocks-cn}:9050";' >/dev/null 2>&1; then
              echo "[SiCABE] CN terdaftar ke FE (attempt $$i)."
              break
            fi
            echo "[SiCABE] Menunggu FE menerima registrasi... (attempt $$i/20)"
            sleep 5
          done
        fi
        bash /opt/starrocks/cn/bin/start_cn.sh
    healthcheck:
      test: ["CMD-SHELL", "mysql -u root -p$${STARROCKS_ROOT_PASSWORD} -h ${STARROCKS_FE_HOST:-starrocks-fe} -P 9030 -e \"SHOW COMPUTE NODES\" 2>/dev/null | grep -q \"Alive\" | head -1"]
      interval: 20s
      timeout: 15s
      retries: 15
      start_period: 120s
    profiles:
      - phase4
      - all


# ============================================================================
# PHASE 4b — Serving (Paralel): Dremio OSS (query engine di atas Nessie)
# Dependency: Storage (MinIO, Nessie). Auth lokal (first-run admin via UI).
# ============================================================================

  dremio:
    image: dremio/dremio-oss:${DREMIO_VERSION:-25.2.0}
    container_name: sicabe-dremio
    restart: unless-stopped
    networks:
      - sicabe-network
    # Tanpa depends_on: Dremio opt-in (profile `dremio`) sedangkan minio/nessie
    # berada di profile `all`. `depends_on` ke service di luar profile aktif
    # membuat Compose menolak project ("depends on undefined service"). Saat
    # diaktifkan, stack utama sudah berjalan sehingga urutan tak diperlukan.
    environment:
      # Catatan: Dremio MENOLAK start bila heap < 4 GB ("Dremio requires at
      # least 4 GB memory to run") → heap-nya TIDAK bisa diturunkan seperti
      # komponen JVM lain. Biarkan default: heap 4096 MB dan
      # MaxDirectMemorySize 8192 MB (yang terakhir hanya plafon, bukan alokasi).
      DREMIO_JAVA_SERVER_EXTRA_OPTS: "-Dpaths.dist=file:///opt/dremio/data/dist"
    volumes:
      - dremio-data:/opt/dremio/data
    ports:
      - "${DREMIO_UI_PORT:-12947}:9047"       # Web UI
      - "${DREMIO_JDBC_PORT:-12010}:31010"    # JDBC (legacy)
      - "${DREMIO_FLIGHT_PORT:-12011}:32010"  # Arrow Flight
    # OPT-IN: Dremio tidak ikut `--profile all` / `phaseN`. Aktifkan khusus:
    #   docker compose --profile dremio up -d dremio
    # Alasan: Dremio menolak heap < 4 GB (hard requirement installer-nya)
    # sehingga memakan RAM besar; di mode dev umumnya tidak diperlukan.
    profiles:
      - dremio

# ============================================================================
# PHASE 5 — Ingestion Layer: Airflow
# Dependency: Storage (MinIO, Nessie) + Processing (Trino) + Keycloak
# ============================================================================

  # ─── Init volume: izin tulis untuk user airflow (uid 50000) ────────────────
  # Volume bernama dibuat Docker sebagai root:root. Plugin OpenMetadata menulis
  # config pipeline sebagai user airflow sehingga GAGAL SENYAP bila izin belum
  # diperbaiki — gejalanya DAG hasil plugin tidak bisa diimpor:
  #   ConfigurationError: Cannot open config file .../dag_generated_configs/<n>.json
  # Container ini berjalan sekali sebagai root, membetulkan izin, lalu keluar.
  airflow-volume-init:
    image: ${AIRFLOW_IMAGE:-sicabe/airflow:3.3.2-om203-ol}
    container_name: sicabe-airflow-volume-init
    restart: "no"
    user: root
    networks:
      - sicabe-network
    entrypoint: ["/bin/bash", "-c"]
    command:
      - "mkdir -p /cfg /dags /logs && chown -R 50000:0 /cfg /dags /logs && chmod 775 /cfg && echo 'Izin volume Airflow siap (uid 50000)'"
    volumes:
      - airflow-dags-configs:/cfg
      - airflow-dags:/dags
      - airflow-logs:/logs
    profiles:
      - phase5
      - all

  airflow-init:
    image: ${AIRFLOW_IMAGE:-sicabe/airflow:3.3.2-om203-ol}
    container_name: sicabe-airflow-init
    # Hostname HARUS bisa diresolusi. Secara default Docker memberi hostname = ID
    # container, dan task log server Airflow memakai socket.gethostname() sebagai
    # alamatnya -> muncul "Could not read served logs: HTTPConnectionPool(host=
    # '<id-container>', port=8793) ... Connection refused" (kosmetik, tapi
    # membingungkan). Dengan hostname = nama container, alamat itu resolvable.
    hostname: sicabe-airflow-init
    restart: "no"
    networks:
      - sicabe-network
    depends_on:
      postgres:
        condition: service_healthy
      # Perbaiki izin volume (root) SEBELUM penulisan DAG/config oleh plugin.
      airflow-volume-init:
        condition: service_completed_successfully
    volumes:
      - ./config/airflow/airflow.cfg:/opt/airflow/airflow.cfg:ro
      - ./config/airflow/simple_auth_manager_passwords.json:/opt/airflow/simple_auth_manager_passwords.json
      - airflow-dags:/opt/airflow/dags
      - airflow-dags-configs:/opt/airflow/dag_generated_configs
      - airflow-logs:/opt/airflow/logs
      - ./scripts/airflow:/opt/airflow/scripts:ro
      - ./config/airflow/dbt:/opt/airflow/dbt
      - ./config/airflow/great_expectations:/opt/airflow/great_expectations:ro
      - ./config/airflow/sling:/opt/airflow/sling:ro
      # Sertifikat Trino (PEM) untuk klien yang MEMVERIFIKASI TLS — dipakai
      # workflow ingestion OpenMetadata lewat field `verify` (dbt memakai cert:false).
      - ./config/trino/certs:/etc/ssl/trino:ro
    environment:
      # dbt menulis artefak ke <project>/target dan log ke <project>/logs. Direktori
      # dbt di-mount dari repo (pemilik uid host) sementara container berjalan sebagai
      # uid 50000 (airflow) -> TIDAK bisa menulis, dan dbt gagal dengan
      #   [Errno 13] Permission denied: '/opt/airflow/dbt/sicabe/target'
      # Bahkan pesannya pun tidak terlihat karena log-nya juga tidak bisa ditulis
      # (gejalanya: "exit code 2" tanpa keluaran sama sekali). Arahkan ke /tmp.
      DBT_LOG_PATH: ${DBT_LOG_PATH:-/tmp/dbt/logs}
      DBT_TARGET_PATH: ${DBT_TARGET_PATH:-/tmp/dbt/target}
      AIRFLOW__DATABASE__SQL_ALCHEMY_CONN: postgresql+psycopg2://airflow:${AIRFLOW_DB_PASSWORD:?err}@${POSTGRES_HOST:-postgres}:${POSTGRES_INTERNAL_PORT:-5432}/airflow
      AIRFLOW__CORE__EXECUTOR: ${AIRFLOW_EXECUTOR:-LocalExecutor}
      # Secret key konsisten untuk JWT internal (task runner auth)
      AIRFLOW__API_AUTH__JWT_SECRET: ${AIRFLOW_SECRET_KEY:?err}
      # SimpleAuthManager users (format: username:password[:team]).
      # ⚠️ Field ketiga adalah TEAM, bukan role — mengisinya memicu crash
      # "user ... is associated to at least one team and multi-team mode is not
      # configured" (dibuktikan di experiments/om20x). Jadi cukup dua field.
      AIRFLOW__CORE__SIMPLE_AUTH_MANAGER_USERS: ${AIRFLOW_SIMPLE_AUTH_USERS:-admin:Admin}
      # Password superuser PostgreSQL (untuk DAG akses sicabe_data)
      POSTGRES_PASSWORD: ${POSTGRES_PASSWORD:?err}
      # ─── Metrik → statsd-exporter (M2 observabilitas, Review 19) ───
      AIRFLOW__METRICS__STATSD_ON: "true"
      AIRFLOW__METRICS__STATSD_HOST: ${STATSD_HOST:-statsd-exporter}
      AIRFLOW__METRICS__STATSD_PORT: ${STATSD_PORT:-9125}
      AIRFLOW__METRICS__STATSD_PREFIX: airflow
    # Note: Airflow 3.x menggunakan 'airflow db migrate' (bukan 'db init')
    # Admin user dibuat via 'airflow users create' (SimpleAuthManager — bukan lagi _AIRFLOW_WWW_USER_*)
    command:
      - bash
      - -c
      - airflow db migrate && airflow users create --username admin --password ${AIRFLOW_ADMIN_PASSWORD:?err} --role Admin --email admin@sicabe.local --firstname Admin --lastname User 2>/dev/null || true && echo 'Airflow DB initialized + admin user'
    profiles:
      - phase5
      - all

  airflow-webserver:
    image: ${AIRFLOW_IMAGE:-sicabe/airflow:3.3.2-om203-ol}
    container_name: sicabe-airflow-webserver
    # Hostname HARUS bisa diresolusi. Secara default Docker memberi hostname = ID
    # container, dan task log server Airflow memakai socket.gethostname() sebagai
    # alamatnya -> muncul "Could not read served logs: HTTPConnectionPool(host=
    # '<id-container>', port=8793) ... Connection refused" (kosmetik, tapi
    # membingungkan). Dengan hostname = nama container, alamat itu resolvable.
    hostname: sicabe-airflow-webserver
    labels:
      - "traefik.enable=true"
      - "traefik.http.routers.airflow.rule=Host(`airflow.${CLUSTER_DOMAIN:-localhost}`)"
      - "traefik.http.routers.airflow.entrypoints=websecure"
      - "traefik.http.routers.airflow.tls=true"
      - "traefik.http.routers.airflow.tls.certresolver=le"
      - "traefik.http.routers.airflow.tls.domains[0].main=${CLUSTER_DOMAIN:-localhost}"
      - "traefik.http.routers.airflow.tls.domains[0].sans=*.${CLUSTER_DOMAIN:-localhost}"
      - "traefik.http.services.airflow.loadbalancer.server.port=8080"
    restart: unless-stopped
    networks:
      - sicabe-network
    depends_on:
      airflow-init:
        condition: service_completed_successfully
      postgres:
        condition: service_healthy
    volumes:
      - ./config/airflow/airflow.cfg:/opt/airflow/airflow.cfg:ro
      - ./config/airflow/simple_auth_manager_passwords.json:/opt/airflow/simple_auth_manager_passwords.json
      - airflow-dags:/opt/airflow/dags
      - airflow-dags-configs:/opt/airflow/dag_generated_configs
      - airflow-logs:/opt/airflow/logs
      - ./scripts/airflow:/opt/airflow/scripts:ro
      - ./config/airflow/dbt:/opt/airflow/dbt
      - ./config/airflow/great_expectations:/opt/airflow/great_expectations:ro
      - ./config/airflow/sling:/opt/airflow/sling:ro
      # Sertifikat Trino (PEM) untuk klien yang MEMVERIFIKASI TLS — dipakai
      # workflow ingestion OpenMetadata lewat field `verify` (dbt memakai cert:false).
      - ./config/trino/certs:/etc/ssl/trino:ro
    environment:
      # dbt menulis artefak ke <project>/target dan log ke <project>/logs. Direktori
      # dbt di-mount dari repo (pemilik uid host) sementara container berjalan sebagai
      # uid 50000 (airflow) -> TIDAK bisa menulis, dan dbt gagal dengan
      #   [Errno 13] Permission denied: '/opt/airflow/dbt/sicabe/target'
      # Bahkan pesannya pun tidak terlihat karena log-nya juga tidak bisa ditulis
      # (gejalanya: "exit code 2" tanpa keluaran sama sekali). Arahkan ke /tmp.
      DBT_LOG_PATH: ${DBT_LOG_PATH:-/tmp/dbt/logs}
      DBT_TARGET_PATH: ${DBT_TARGET_PATH:-/tmp/dbt/target}
      AIRFLOW__DATABASE__SQL_ALCHEMY_CONN: postgresql+psycopg2://airflow:${AIRFLOW_DB_PASSWORD:?err}@${POSTGRES_HOST:-postgres}:${POSTGRES_INTERNAL_PORT:-5432}/airflow
      AIRFLOW__CORE__EXECUTOR: ${AIRFLOW_EXECUTOR:-LocalExecutor}
      # URL browser-facing (link di UI). Default localhost, bukan nama internal
      # airflow-webserver:8080 yang tidak bisa diresolusi browser.
      AIRFLOW__WEBSERVER__BASE_URL: ${AIRFLOW_PUBLIC_URL:-http://localhost:${AIRFLOW_PORT:-12082}}
      # Secret key konsisten untuk JWT internal (task runner auth)
      AIRFLOW__API_AUTH__JWT_SECRET: ${AIRFLOW_SECRET_KEY:?err}
      # SimpleAuthManager users (format: username:password[:team]).
      # ⚠️ Field ketiga adalah TEAM, bukan role — mengisinya memicu crash
      # "user ... is associated to at least one team and multi-team mode is not
      # configured" (dibuktikan di experiments/om20x). Jadi cukup dua field.
      AIRFLOW__CORE__SIMPLE_AUTH_MANAGER_USERS: ${AIRFLOW_SIMPLE_AUTH_USERS:-admin:Admin}
      # Password superuser PostgreSQL (untuk DAG akses sicabe_data)
      POSTGRES_PASSWORD: ${POSTGRES_PASSWORD:?err}
      # Token OpenMetadata untuk GE Action (validate_to_openmetadata.py) — dipakai
      # task validasi DAG agar hasil Great Expectations TIDAK berhenti di log,
      # tetapi terkirim ke OpenMetadata (Data Quality Test Suite/Case).
      OPENMETADATA_ADMIN_PASSWORD: ${OPENMETADATA_ADMIN_PASSWORD:?err}
      OM_API_URL: http://${OPENMETADATA_HOST:-openmetadata-server}:8585/api
      # dbt → Trino: service account via HTTPS 8443 langsung (self-signed, cert: false)
      DBT_TRINO_HOST: ${DBT_TRINO_HOST:-trino-coordinator}
      DBT_TRINO_USER: ${DBT_TRINO_USER:-svc_dbt}
      DBT_TRINO_PASSWORD: ${TRINO_SVC_DBT_PASSWORD:?err}
      # ─── OpenLineage → OpenMetadata (experiments/lineage-airflow-om) ───
      # Transport dibaca dari KONEKSI Airflow (`config_conn_id`) — token TIDAK
      # ditulis di .env. Koneksi dibuat bootstrap.sh (`openlineage_config` +
      # `openmetadata_lineage_api`). Kosong = provider INERT.
      AIRFLOW__OPENLINEAGE__CONFIG_CONN_ID: ${AIRFLOW_OPENLINEAGE_CONFIG_CONN_ID:-}
      AIRFLOW__OPENLINEAGE__NAMESPACE: ${AIRFLOW_OPENLINEAGE_NAMESPACE:-}
      AIRFLOW__OPENLINEAGE__EXECUTE_IN_THREAD: "true"
      # ─── Metrik → statsd-exporter (M2 observabilitas, Review 19) ───
      AIRFLOW__METRICS__STATSD_ON: "true"
      AIRFLOW__METRICS__STATSD_HOST: ${STATSD_HOST:-statsd-exporter}
      AIRFLOW__METRICS__STATSD_PORT: ${STATSD_PORT:-9125}
      AIRFLOW__METRICS__STATSD_PREFIX: airflow
      # Catatan: Airflow 3.3 memakai SimpleAuthManager (bukan FabAuthManager),
      # jadi _AIRFLOW_WWW_USER_* TIDAK dipakai dan memicu crash di entrypoint.
      # Admin user dibuat oleh airflow-init via 'airflow users create'.
    # Airflow 3.x: 'airflow webserver' diganti 'airflow api-server'
    # Health endpoint: /api/v2/monitor/health (bukan /health)
    # Entrypoint wrapper: install sling CLI dulu, lalu start service
    command: ["bash", "/opt/airflow/scripts/entrypoint.sh", "airflow", "api-server"]
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8080/api/v2/monitor/health"]
      interval: 15s
      timeout: 10s
      retries: 10
      start_period: 30s
    ports:
      - "${AIRFLOW_PORT:-12082}:8080"
    profiles:
      - phase5
      - all

  airflow-dag-processor:
    image: ${AIRFLOW_IMAGE:-sicabe/airflow:3.3.2-om203-ol}
    container_name: sicabe-airflow-dag-processor
    # Hostname HARUS bisa diresolusi. Secara default Docker memberi hostname = ID
    # container, dan task log server Airflow memakai socket.gethostname() sebagai
    # alamatnya -> muncul "Could not read served logs: HTTPConnectionPool(host=
    # '<id-container>', port=8793) ... Connection refused" (kosmetik, tapi
    # membingungkan). Dengan hostname = nama container, alamat itu resolvable.
    hostname: sicabe-airflow-dag-processor
    restart: unless-stopped
    networks:
      - sicabe-network
    depends_on:
      airflow-init:
        condition: service_completed_successfully
      postgres:
        condition: service_healthy
    volumes:
      - ./config/airflow/airflow.cfg:/opt/airflow/airflow.cfg:ro
      - ./config/airflow/simple_auth_manager_passwords.json:/opt/airflow/simple_auth_manager_passwords.json
      - airflow-dags:/opt/airflow/dags
      - airflow-dags-configs:/opt/airflow/dag_generated_configs
      - airflow-logs:/opt/airflow/logs
      - ./scripts/airflow:/opt/airflow/scripts:ro
      - ./config/airflow/dbt:/opt/airflow/dbt
      - ./config/airflow/great_expectations:/opt/airflow/great_expectations:ro
      - ./config/airflow/sling:/opt/airflow/sling:ro
      # Sertifikat Trino (PEM) untuk klien yang MEMVERIFIKASI TLS — dipakai
      # workflow ingestion OpenMetadata lewat field `verify` (dbt memakai cert:false).
      - ./config/trino/certs:/etc/ssl/trino:ro
    environment:
      # dbt menulis artefak ke <project>/target dan log ke <project>/logs. Direktori
      # dbt di-mount dari repo (pemilik uid host) sementara container berjalan sebagai
      # uid 50000 (airflow) -> TIDAK bisa menulis, dan dbt gagal dengan
      #   [Errno 13] Permission denied: '/opt/airflow/dbt/sicabe/target'
      # Bahkan pesannya pun tidak terlihat karena log-nya juga tidak bisa ditulis
      # (gejalanya: "exit code 2" tanpa keluaran sama sekali). Arahkan ke /tmp.
      DBT_LOG_PATH: ${DBT_LOG_PATH:-/tmp/dbt/logs}
      DBT_TARGET_PATH: ${DBT_TARGET_PATH:-/tmp/dbt/target}
      AIRFLOW__DATABASE__SQL_ALCHEMY_CONN: postgresql+psycopg2://airflow:${AIRFLOW_DB_PASSWORD:?err}@${POSTGRES_HOST:-postgres}:${POSTGRES_INTERNAL_PORT:-5432}/airflow
      AIRFLOW__CORE__EXECUTOR: ${AIRFLOW_EXECUTOR:-LocalExecutor}
      # Secret key konsisten untuk JWT internal (task runner auth)
      AIRFLOW__API_AUTH__JWT_SECRET: ${AIRFLOW_SECRET_KEY:?err}
      # SimpleAuthManager users (format: username:password[:team]).
      # ⚠️ Field ketiga adalah TEAM, bukan role — mengisinya memicu crash
      # "user ... is associated to at least one team and multi-team mode is not
      # configured" (dibuktikan di experiments/om20x). Jadi cukup dua field.
      AIRFLOW__CORE__SIMPLE_AUTH_MANAGER_USERS: ${AIRFLOW_SIMPLE_AUTH_USERS:-admin:Admin}
      # Password superuser PostgreSQL (untuk DAG akses sicabe_data)
      POSTGRES_PASSWORD: ${POSTGRES_PASSWORD:?err}
      # Token OpenMetadata untuk GE Action (validate_to_openmetadata.py) — dipakai
      # task validasi DAG agar hasil Great Expectations TIDAK berhenti di log,
      # tetapi terkirim ke OpenMetadata (Data Quality Test Suite/Case).
      OPENMETADATA_ADMIN_PASSWORD: ${OPENMETADATA_ADMIN_PASSWORD:?err}
      OM_API_URL: http://${OPENMETADATA_HOST:-openmetadata-server}:8585/api
      # dbt → Trino: service account via HTTPS 8443 langsung (self-signed, cert: false)
      DBT_TRINO_HOST: ${DBT_TRINO_HOST:-trino-coordinator}
      DBT_TRINO_USER: ${DBT_TRINO_USER:-svc_dbt}
      DBT_TRINO_PASSWORD: ${TRINO_SVC_DBT_PASSWORD:?err}
      # ─── OpenLineage → OpenMetadata (experiments/lineage-airflow-om) ───
      # Transport dibaca dari KONEKSI Airflow (`config_conn_id`) — token TIDAK
      # ditulis di .env. Koneksi dibuat bootstrap.sh (`openlineage_config` +
      # `openmetadata_lineage_api`). Kosong = provider INERT.
      AIRFLOW__OPENLINEAGE__CONFIG_CONN_ID: ${AIRFLOW_OPENLINEAGE_CONFIG_CONN_ID:-}
      AIRFLOW__OPENLINEAGE__NAMESPACE: ${AIRFLOW_OPENLINEAGE_NAMESPACE:-}
      AIRFLOW__OPENLINEAGE__EXECUTE_IN_THREAD: "true"
      # ─── Metrik → statsd-exporter (M2 observabilitas, Review 19) ───
      AIRFLOW__METRICS__STATSD_ON: "true"
      AIRFLOW__METRICS__STATSD_HOST: ${STATSD_HOST:-statsd-exporter}
      AIRFLOW__METRICS__STATSD_PORT: ${STATSD_PORT:-9125}
      AIRFLOW__METRICS__STATSD_PREFIX: airflow
    # Airflow 3.x: DAG parsing dilakukan oleh dag-processor (pisah dari scheduler)
    command: ["bash", "/opt/airflow/scripts/entrypoint.sh", "airflow", "dag-processor"]
    healthcheck:
      # Image Airflow minimal tidak punya pgrep/ps — scan /proc via python
      test: ["CMD-SHELL", "python3 -c \"import os; print('ALIVE' if any('dag-processor' in open(f'/proc/{p}/cmdline').read() for p in os.listdir('/proc') if p.isdigit() and os.path.exists(f'/proc/{p}/cmdline')) else 'DEAD')\" | grep -q ALIVE"]
      interval: 15s
      timeout: 10s
      retries: 10
      start_period: 60s
    profiles:
      - phase5
      - all

  airflow-scheduler:
    image: ${AIRFLOW_IMAGE:-sicabe/airflow:3.3.2-om203-ol}
    container_name: sicabe-airflow-scheduler
    # Hostname HARUS bisa diresolusi. Secara default Docker memberi hostname = ID
    # container, dan task log server Airflow memakai socket.gethostname() sebagai
    # alamatnya -> muncul "Could not read served logs: HTTPConnectionPool(host=
    # '<id-container>', port=8793) ... Connection refused" (kosmetik, tapi
    # membingungkan). Dengan hostname = nama container, alamat itu resolvable.
    hostname: sicabe-airflow-scheduler
    restart: unless-stopped
    networks:
      - sicabe-network
    depends_on:
      airflow-dag-processor:
        condition: service_started
      airflow-init:
        condition: service_completed_successfully
      postgres:
        condition: service_healthy
    volumes:
      - ./config/airflow/airflow.cfg:/opt/airflow/airflow.cfg:ro
      - ./config/airflow/simple_auth_manager_passwords.json:/opt/airflow/simple_auth_manager_passwords.json
      - airflow-dags:/opt/airflow/dags
      - airflow-dags-configs:/opt/airflow/dag_generated_configs
      - airflow-logs:/opt/airflow/logs
      - ./scripts/airflow:/opt/airflow/scripts:ro
      - ./config/airflow/dbt:/opt/airflow/dbt
      - ./config/airflow/great_expectations:/opt/airflow/great_expectations:ro
      - ./config/airflow/sling:/opt/airflow/sling:ro
      # Sertifikat Trino (PEM) untuk klien yang MEMVERIFIKASI TLS — dipakai
      # workflow ingestion OpenMetadata lewat field `verify` (dbt memakai cert:false).
      - ./config/trino/certs:/etc/ssl/trino:ro
    environment:
      # dbt menulis artefak ke <project>/target dan log ke <project>/logs. Direktori
      # dbt di-mount dari repo (pemilik uid host) sementara container berjalan sebagai
      # uid 50000 (airflow) -> TIDAK bisa menulis, dan dbt gagal dengan
      #   [Errno 13] Permission denied: '/opt/airflow/dbt/sicabe/target'
      # Bahkan pesannya pun tidak terlihat karena log-nya juga tidak bisa ditulis
      # (gejalanya: "exit code 2" tanpa keluaran sama sekali). Arahkan ke /tmp.
      DBT_LOG_PATH: ${DBT_LOG_PATH:-/tmp/dbt/logs}
      DBT_TARGET_PATH: ${DBT_TARGET_PATH:-/tmp/dbt/target}
      AIRFLOW__DATABASE__SQL_ALCHEMY_CONN: postgresql+psycopg2://airflow:${AIRFLOW_DB_PASSWORD:?err}@${POSTGRES_HOST:-postgres}:${POSTGRES_INTERNAL_PORT:-5432}/airflow
      AIRFLOW__CORE__EXECUTOR: ${AIRFLOW_EXECUTOR:-LocalExecutor}
      # Secret key konsisten untuk JWT internal (task runner auth)
      AIRFLOW__API_AUTH__JWT_SECRET: ${AIRFLOW_SECRET_KEY:?err}
      # SimpleAuthManager users (format: username:password[:team]).
      # ⚠️ Field ketiga adalah TEAM, bukan role — mengisinya memicu crash
      # "user ... is associated to at least one team and multi-team mode is not
      # configured" (dibuktikan di experiments/om20x). Jadi cukup dua field.
      AIRFLOW__CORE__SIMPLE_AUTH_MANAGER_USERS: ${AIRFLOW_SIMPLE_AUTH_USERS:-admin:Admin}
      # Password superuser PostgreSQL (untuk DAG akses sicabe_data)
      POSTGRES_PASSWORD: ${POSTGRES_PASSWORD:?err}
      # Token OpenMetadata untuk GE Action (validate_to_openmetadata.py) — dipakai
      # task validasi DAG agar hasil Great Expectations TIDAK berhenti di log,
      # tetapi terkirim ke OpenMetadata (Data Quality Test Suite/Case).
      OPENMETADATA_ADMIN_PASSWORD: ${OPENMETADATA_ADMIN_PASSWORD:?err}
      OM_API_URL: http://${OPENMETADATA_HOST:-openmetadata-server}:8585/api
      # dbt → Trino: service account via HTTPS 8443 langsung (self-signed, cert: false)
      DBT_TRINO_HOST: ${DBT_TRINO_HOST:-trino-coordinator}
      DBT_TRINO_USER: ${DBT_TRINO_USER:-svc_dbt}
      DBT_TRINO_PASSWORD: ${TRINO_SVC_DBT_PASSWORD:?err}
      # ─── OpenLineage → OpenMetadata (experiments/lineage-airflow-om) ───
      # Konfigurasi transport dibaca dari KONEKSI Airflow (`config_conn_id`),
      # BUKAN dari env — supaya token lineage TIDAK perlu ditulis di .env.
      # Koneksi dibuat oleh bootstrap.sh:
      #   - `openlineage_config`       : extra = {"transport": {...}} dengan
      #                                  auth type `airflow_connection_api_key`
      #   - `openmetadata_lineage_api` : password = token bot `lineage-bot`
      # ⚠️ Pakai token BOT `lineage-bot` (role LineageBotRole, bisa UNLIMITED).
      #    Token admin hanya bertahan 1 JAM → tidak layak untuk konfigurasi.
      # Kosong (default) = provider INERT: tidak mengirim apa pun.
      AIRFLOW__OPENLINEAGE__CONFIG_CONN_ID: ${AIRFLOW_OPENLINEAGE_CONFIG_CONN_ID:-}
      AIRFLOW__OPENLINEAGE__NAMESPACE: ${AIRFLOW_OPENLINEAGE_NAMESPACE:-}
      # WAJIB saat aktif: tanpa ini proses emisi HANG lalu dimatikan setelah
      # `[openlineage] execution_timeout`, sehingga TIDAK ADA event keluar
      # (lihat experiments/lineage-airflow-om/FASE4-HASIL.md).
      AIRFLOW__OPENLINEAGE__EXECUTE_IN_THREAD: "true"
      # ─── Metrik → statsd-exporter (M2 observabilitas, Review 19) ───
      AIRFLOW__METRICS__STATSD_ON: "true"
      AIRFLOW__METRICS__STATSD_HOST: ${STATSD_HOST:-statsd-exporter}
      AIRFLOW__METRICS__STATSD_PORT: ${STATSD_PORT:-9125}
      AIRFLOW__METRICS__STATSD_PREFIX: airflow
    command: ["bash", "/opt/airflow/scripts/entrypoint.sh", "airflow", "scheduler"]
    healthcheck:
      # Cek scheduler masih alive via airflow jobs check (exit 0 = ada job hidup)
      test: ["CMD-SHELL", "airflow jobs check > /dev/null 2>&1 || exit 1"]
      interval: 15s
      timeout: 10s
      retries: 10
      start_period: 60s
    profiles:
      - phase5
      - all


# ============================================================================
# PHASE 6 — Governance Layer (Lanjutan): OpenMetadata → Grafana
# Dependency: Semua layer sebelumnya sudah running
# ============================================================================

  openmetadata-opensearch:
    # WAJIB OpenSearch 3.x untuk OpenMetadata 2.0.x (minimum 3.0.0).
    # ⚠️ Hindari 3.3.2: bug Lucene 10.3.1 membuat pencarian mengembalikan HTTP 200
    #    dengan 0 hasil (shard gagal senyap) dan OM tidak melaporkannya. 3.4.0+ benar.
    # ⚠️ Index 2.19.6 TIDAK bisa dibaca 3.x — index ini derived, OM membangun ulang.
    image: opensearchproject/opensearch:3.8.0
    container_name: sicabe-opensearch
    restart: unless-stopped
    networks:
      - sicabe-network
    environment:
      discovery.type: single-node
      # OpenSearch 3.x memakai env ini; `plugins.security.disabled` (gaya 2.x)
      # tidak lagi dipakai.
      DISABLE_SECURITY_PLUGIN: "true"
      bootstrap.memory_lock: "true"
      OPENSEARCH_JAVA_OPTS: "-Xms512m -Xmx512m"
      # OPENSEARCH_INITIAL_ADMIN_PASSWORD dihapus: tidak dipakai saat security
      # plugin dimatikan, dan menaruh kredensial di compose melanggar aturan
      # "jangan hardcode kredensial".
    ulimits:
      memlock:
        soft: -1
        hard: -1
      nofile:
        soft: 65536
        hard: 65536
    volumes:
      - opensearch-data:/usr/share/opensearch/data
    healthcheck:
      test: ["CMD-SHELL", "curl -fs http://localhost:9200/_cluster/health > /dev/null || exit 1"]
      interval: 15s
      timeout: 10s
      retries: 12
      start_period: 45s
    ports:
      - "${OPENSEARCH_PORT:-12920}:9200"
    profiles:
      - phase6
      - all

  # ─── One-shot: migrasi schema OpenMetadata (idempotent) ───────────────────
  # OM tidak auto-migrate saat server start → migrasi harus dijalankan lebih
  # dulu, jika tidak server crash-loop (tabel belum ada). OM 2.0.x memakai
  # perintah yang sama seperti 1.13.x.
  openmetadata-migrate:
    image: openmetadata/server:${OPENMETADATA_VERSION:-2.0.3}
    container_name: sicabe-openmetadata-migrate
    restart: "no"
    networks:
      - sicabe-network
    depends_on:
      postgres:
        condition: service_healthy
    entrypoint: ["/bin/sh", "-c"]
    command:
      - "cd /opt/openmetadata && ./bootstrap/openmetadata-ops.sh -d migrate --force"
    environment:
      DB_DRIVER_CLASS: org.postgresql.Driver
      DB_HOST: ${POSTGRES_HOST:-postgres}
      DB_PORT: 5432
      DB_DATABASE: openmetadata_db
      DB_USER: openmetadata
      DB_PASSWORD: ${OPENMETADATA_DB_PASSWORD:?err}
      DB_USER_PASSWORD: ${OPENMETADATA_DB_PASSWORD:?err}
      DB_SCHEME: postgresql
    profiles:
      - phase6
      - all

  openmetadata-server:
    image: openmetadata/server:${OPENMETADATA_VERSION:-2.0.3}
    container_name: sicabe-openmetadata
    labels:
      - "traefik.enable=true"
      - "traefik.http.routers.om.rule=Host(`om.${CLUSTER_DOMAIN:-localhost}`)"
      - "traefik.http.routers.om.entrypoints=websecure"
      - "traefik.http.routers.om.tls=true"
      - "traefik.http.routers.om.tls.certresolver=le"
      - "traefik.http.routers.om.tls.domains[0].main=${CLUSTER_DOMAIN:-localhost}"
      - "traefik.http.routers.om.tls.domains[0].sans=*.${CLUSTER_DOMAIN:-localhost}"
      - "traefik.http.services.om.loadbalancer.server.port=8585"
    restart: unless-stopped
    networks:
      - sicabe-network
    depends_on:
      openmetadata-opensearch:
        condition: service_healthy
      postgres:
        condition: service_healthy
      openmetadata-migrate:
        condition: service_completed_successfully
    volumes:
      - openmetadata-data:/data
    environment:
      DB_DRIVER_CLASS: org.postgresql.Driver
      DB_HOST: ${POSTGRES_HOST:-postgres}
      DB_PORT: 5432
      DB_DATABASE: openmetadata_db
      DB_USER: openmetadata
      DB_PASSWORD: ${OPENMETADATA_DB_PASSWORD:?err}
      # OM 2.0.x membaca DB_PASSWORD; DB_USER_PASSWORD dipertahankan karena
      # sebagian rilis/tooling masih membacanya (migrasi & ops script).
      DB_USER_PASSWORD: ${OPENMETADATA_DB_PASSWORD:?err}
      # DB_SCHEME wajib di-override: default OM adalah mysql.
      DB_SCHEME: postgresql
      # Elasticsearch/OpenSearch connection
      ELASTICSEARCH_HOST: ${OPENSEARCH_HOST:-openmetadata-opensearch}
      ELASTICSEARCH_PORT: 9200
      SEARCH_TYPE: opensearch
      # OM membaca SERVER_PORT (bukan OPENMETADATA_SERVER_PORT); pakai port
      # internal 8585 (bukan port host) agar konsisten dgn mapping 12085:8585.
      SERVER_PORT: 8585
      # Endpoint API OM yang DITULIS OM ke config pipeline yang dibangkitkannya
      # (openmetadata.yaml: metadataApiEndpoint = ${SERVER_HOST_API_URL:-localhost:8585/api}).
      # Default localhost TIDAK bisa dijangkau dari container Airflow -> DAG hasil
      # plugin gagal: 'Transport failure calling http://localhost:8585/api/...'.
      # Karena ingestion/DQ berjalan DI DALAM jaringan compose, pakai nama service.
      SERVER_HOST_API_URL: http://${OPENMETADATA_HOST:-openmetadata-server}:8585/api
      # CATATAN: SERVER_HOST_API_URL sengaja TETAP internal — ia dipakai sisi SERVER
      # (pipelineServiceClientConfiguration.metadataApiEndpoint) oleh Airflow/ingestion
      # untuk memanggil balik OM. Kalau diarahkan ke URL publik, ingestion justru rusak
      # bila nama itu tidak bisa diresolusi container.
      #
      # OPENMETADATA_SERVER_URL = URL yang dipakai MEMBANGUN LINK di email/notifikasi
      # (browser-facing). OM tidak punya env var "UI base URL": UI memakai path relatif,
      # jadi browser memakai origin tempat ia dimuat — tidak perlu diatur.
      OPENMETADATA_SERVER_URL: ${OPENMETADATA_PUBLIC_URL:-http://localhost:${OPENMETADATA_PORT:-12085}}
      # OIDC SSO via Keycloak (bab 4.6.1.5) — set ke 'basic' untuk local login dulu
      AUTHENTICATION_PROVIDER: basic
      # OIDC_CLIENT_ID: openmetadata
      # OIDC_CLIENT_SECRET: "${OPENMETADATA_OIDC_SECRET:-om-secret}"
      # OIDC_DISCOVERY_URI: http://keycloak:8080/realms/${SICABE_REALM:-sicabe}/.well-known/openid-configuration
      # OIDC_SERVER_URL: http://keycloak:8080/realms/${SICABE_REALM:-sicabe}
      # Admin user bootstrap
      OPENMETADATA_ADMIN_PASSWORD: ${OPENMETADATA_ADMIN_PASSWORD:-openmetadata123}
      # ─── Pipeline Service Client (Airflow) ──────────────────────────────────
      # Tanpa ini OM menunjuk http://localhost:8080 (di dalam container = kosong),
      # sehingga tombol "Test Connection" di UI DISABLED dengan tooltip
      # "Platform Service Client Unavailable" — pesan itu label UI, bukan plugin
      # yang kurang, dan bukan spesifik Trino. Endpoint + kredensial Airflow
      # (SimpleAuthManager: admin + AIRFLOW_ADMIN_PASSWORD) membuat tombol aktif
      # sekaligus memungkinkan UI menjalankan ingestion/DQ pipeline lewat Airflow.
      PIPELINE_SERVICE_CLIENT_ENABLED: "true"
      PIPELINE_SERVICE_CLIENT_ENDPOINT: http://${AIRFLOW_WEBSERVER_HOST:-airflow-webserver}:8080
      AIRFLOW_USERNAME: admin
      AIRFLOW_PASSWORD: ${AIRFLOW_ADMIN_PASSWORD:?err}
    healthcheck:
      # Image OM tidak menyediakan curl, pakai wget
      # Healthcheck HARUS pakai port internal container (8585), bukan
      # OPENMETADATA_PORT (12085 = port host) — jika tidak, selalu unhealthy.
      test: ["CMD-SHELL", "wget -q -O - http://localhost:8585/ > /dev/null 2>&1 || exit 1"]
      interval: 20s
      timeout: 15s
      retries: 15
      start_period: 120s
    ports:
      - "${OPENMETADATA_PORT:-12085}:8585"
    profiles:
      - phase6
      - all

  # ─── statsd-exporter: metrik Airflow (StatsD) → Prometheus ────────────────
  # Airflow 3 mengirim metrik lewat protokol StatsD (UDP). statsd-exporter
  # mengubahnya ke format Prometheus di :9102, yang lalu di-scrape job `airflow`.
  # Ditempatkan di tier `governance` (bersama Prometheus/Grafana) — pada blueprint
  # a3/b5 tier itu se-host dengan `pipeline` (Airflow), sehingga UDP Airflow→statsd
  # tetap lokal. Lihat lib/cluster/inventory.py (TIER_SERVICES).
  #
  # ✅ PENGIRIM: paket Python `statsd` kini BAKED-IN di image Airflow (ditambahkan ke
  # requirements.332.in + image di-rebuild 2026-10-06). Sebelumnya image belum
  # memuatnya sehingga Airflow memakai NoStatsLogger dan tidak ada metrik terkirim.
  statsd-exporter:
    image: prom/statsd-exporter:${STATSD_EXPORTER_VERSION:-v0.28.0}
    container_name: sicabe-statsd-exporter
    restart: unless-stopped
    networks:
      - sicabe-network
    # 9125/udp = listener StatsD (internal, dipakai Airflow); 9102 = metrik Prometheus.
    ports:
      - "${STATSD_EXPORTER_PORT:-12091}:9102"
    profiles:
      - phase6
      - all

  # ─── Prometheus: sumber metrik untuk datasource Grafana ───────────────────
  # Grafana TIDAK dapat membaca metrik langsung dari MinIO (MinIO = API S3,
  # bukan server Prometheus). Prometheus ini yang meng-scrape metrik MinIO &
  # StarRocks, lalu Grafana membacanya dari sini.
  prometheus:
    image: prom/prometheus:${PROMETHEUS_VERSION:-v3.6.0}
    container_name: sicabe-prometheus
    restart: unless-stopped
    networks:
      - sicabe-network
    extra_hosts:
      # `host.docker.internal` menunjuk Docker host itu sendiri — dipakai target
      # default node.yml. `host-gateway` tersedia sejak Docker 20.10.
      - "host.docker.internal:host-gateway"
    volumes:
      - ./config/prometheus/prometheus.yml:/etc/prometheus/prometheus.yml:ro
      # Daftar target node_exporter (file_sd) — dapat diubah tanpa restart Prometheus.
      - ./config/prometheus/targets:/etc/prometheus/targets:ro
      - prometheus-data:/prometheus
    healthcheck:
      test: ["CMD-SHELL", "wget -q -O - http://localhost:9090/-/healthy >/dev/null 2>&1 || exit 1"]
      interval: 15s
      timeout: 10s
      retries: 10
      start_period: 30s
    ports:
      - "${PROMETHEUS_PORT:-12090}:9090"
    profiles:
      - phase6
      - all

  grafana:
    image: grafana/grafana:${GRAFANA_VERSION:-13.1}
    container_name: sicabe-grafana
    labels:
      - "traefik.enable=true"
      - "traefik.http.routers.grafana.rule=Host(`grafana.${CLUSTER_DOMAIN:-localhost}`)"
      - "traefik.http.routers.grafana.entrypoints=websecure"
      - "traefik.http.routers.grafana.tls=true"
      - "traefik.http.routers.grafana.tls.certresolver=le"
      - "traefik.http.routers.grafana.tls.domains[0].main=${CLUSTER_DOMAIN:-localhost}"
      - "traefik.http.routers.grafana.tls.domains[0].sans=*.${CLUSTER_DOMAIN:-localhost}"
      - "traefik.http.services.grafana.loadbalancer.server.port=3000"
    restart: unless-stopped
    networks:
      - sicabe-network
    volumes:
      - grafana-data:/var/lib/grafana
      - ./config/grafana/datasources:/etc/grafana/provisioning/datasources:ro
      - ./config/grafana/dashboards:/etc/grafana/provisioning/dashboards:ro
    environment:
      GF_SECURITY_ADMIN_USER: admin
      GF_SECURITY_ADMIN_PASSWORD: ${GRAFANA_ADMIN_PASSWORD:-grafana123}
      # Root URL harus pakai host port (bukan port internal 3000) agar
      # redirect_uri OIDC benar saat GRAFANA_PORT != 3000
      GF_SERVER_ROOT_URL: ${GRAFANA_PUBLIC_URL:-http://localhost:${GRAFANA_PORT:-12300}}
      GF_AUTH_GENERIC_OAUTH_ENABLED: "true"
      GF_AUTH_GENERIC_OAUTH_NAME: Keycloak
      GF_AUTH_GENERIC_OAUTH_CLIENT_ID: grafana
      GF_AUTH_GENERIC_OAUTH_CLIENT_SECRET: "${GRAFANA_CLIENT_SECRET:?err}"
      GF_AUTH_GENERIC_OAUTH_AUTH_URL: http://localhost:${KEYCLOAK_PORT:-12080}/realms/${SICABE_REALM:-sicabe}/protocol/openid-connect/auth
      GF_AUTH_GENERIC_OAUTH_TOKEN_URL: http://${KEYCLOAK_HOST:-keycloak}:${KEYCLOAK_INTERNAL_PORT:-8080}/realms/${SICABE_REALM:-sicabe}/protocol/openid-connect/token
      GF_AUTH_GENERIC_OAUTH_API_URL: http://${KEYCLOAK_HOST:-keycloak}:${KEYCLOAK_INTERNAL_PORT:-8080}/realms/${SICABE_REALM:-sicabe}/protocol/openid-connect/userinfo
      GF_AUTH_GENERIC_OAUTH_SCOPES: openid profile email
      GF_AUTH_GENERIC_OAUTH_ROLE_ATTRIBUTE_PATH: contains(roles[*], 'admin') && 'Admin' || 'Viewer'
      GF_AUTH_DISABLE_LOGIN_FORM: "false"
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:3000/api/health"]
      interval: 15s
      timeout: 10s
      retries: 10
      start_period: 30s
    ports:
      - "${GRAFANA_PORT:-12300}:3000"
    profiles:
      - phase6
      - all
  # ── Ingress (T-20, langkah 2) — hanya pada profile `edge` ────────────────────
  # Rute diambil dari LABEL container (provider docker, exposedByDefault=false),
  # jadi daftar rute = label pada layanan; tidak ada konfigurasi rute terpisah.
  # TLS: sertifikat wildcard *.${CLUSTER_DOMAIN} via Let's Encrypt DNS-01.
  # Tanpa kredensial DNS, Traefik TETAP start dan memakai sertifikat self-signed
  # bawaan — jadi routing masih bisa diuji sebelum sertifikat terbit.
  traefik:
    image: traefik:${TRAEFIK_VERSION:-v3.7}
    container_name: sicabe-traefik
    restart: unless-stopped
    profiles:
      - edge
    networks:
      - sicabe-network
    command:
      - "--providers.docker=true"
      - "--providers.docker.exposedbydefault=false"
      - "--providers.file.directory=/etc/traefik/dynamic"
      - "--providers.file.watch=true"
      - "--entrypoints.web.address=:80"
      - "--entrypoints.web.http.redirections.entrypoint.to=websecure"
      - "--entrypoints.web.http.redirections.entrypoint.scheme=https"
      - "--entrypoints.web.http.redirections.entrypoint.permanent=true"
      - "--entrypoints.websecure.address=:443"
      - "--certificatesresolvers.le.acme.email=${TRAEFIK_ACME_EMAIL:-admin@${CLUSTER_DOMAIN:-sicabe.com}}"
      - "--certificatesresolvers.le.acme.storage=/acme/acme.json"
      - "--certificatesresolvers.le.acme.dnschallenge.provider=${TRAEFIK_DNS_PROVIDER:-cloudflare}"
      # Tunggu DNS menyebar sebelum Let's Encrypt memeriksa. Default Traefik 0s — dan
      # untuk DNS-01 dengan nama yang sama (apex + wildcard sama-sama menulis
      # _acme-challenge.<domain>) itu terlalu cepat: LE bisa membaca record lama.
      # Nama opsi yang benar adalah propagation.delayBeforeChecks.
      - "--certificatesresolvers.le.acme.dnschallenge.propagation.delayBeforeChecks=${TRAEFIK_ACME_PROPAGATION_DELAY:-30s}"
      - "--certificatesresolvers.le.acme.caserver=${TRAEFIK_ACME_CASERVER:-https://acme-staging-v02.api.letsencrypt.org/directory}"
    environment:
      CF_DNS_API_TOKEN: ${CF_DNS_API_TOKEN:-}
    ports:
      - "${TRAEFIK_BIND_ADDR:-0.0.0.0}:${TRAEFIK_HTTP_PORT:-80}:80"
      - "${TRAEFIK_BIND_ADDR:-0.0.0.0}:${TRAEFIK_HTTPS_PORT:-443}:443"
    volumes:
      - /var/run/docker.sock:/var/run/docker.sock:ro
      # Named volume (bukan bind mount) supaya acme.json dibuat 0600 oleh Traefik:
      # berkas itu WAJIB mode 600, kalau tidak Traefik menolak start.
      - traefik-acme:/acme
      # serversTransports (di atas) tidak bisa lewat label -> berkas dinamis.
      - ./config/traefik/dynamic.yml:/etc/traefik/dynamic/dynamic.yml:ro